Simple Analogical Reasoning using Word Occurences for Question Answering Passage Retrieval

Toba, Hapnes and Manurung, Ruli and Adriani, Mirna (2010) Simple Analogical Reasoning using Word Occurences for Question Answering Passage Retrieval. Philippine Computing Journal, 5 (2). pp. 1-5. ISSN 1908-1995

[img]
Preview
Text
1.pdf - Published Version

Download (276Kb) | Preview

Abstract

Recent approaches to question answering (QA) attempt to model the relation between existing question-answer pairs, and apply this model to the construction of answers to novel questions. In this paper we study whether a question and its answer can be related using simple word occurence features, and whether this relational model can improve the passage retrieval task of a QA pipeline. We argue that in this context, words appearing in answers in analogous question-answer pairs may represent the information needs of query, thus may also appear in other passages which have some analogical or related features. We attempt to leverage this through query expansion strategies. In our experimental setting, analogy is measured as the similarity between word occurences of related question and answer pairs, and is modeled using the Bayesian Analogical Reasoning (BAR) framework. The resulting model is used to rank retrieved candidate answer passages. Experiments using the ResPubliQA 2009 and 2010 collections show that the analogy-based query expansion does not perform better than the baseline method, but may suggest better performance using more sophisticated linguistic features.

Item Type: Article
Uncontrolled Keywords: Bayesian Analogical Reasoning, ResPubliQA, Mean Reciprocal Rank, Passage Retrieval
Subjects: T Technology > T Technology (General)
Depositing User: Perpustakaan Maranatha
Date Deposited: 20 Dec 2017 07:59
Last Modified: 10 Oct 2018 08:42
URI: http://repository.maranatha.edu/id/eprint/23796

Actions (login required)

View Item View Item